Re: [PATCH 2/2] obstack.c: Fix some sparse warnings

From: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Date: Sep 13, 2011, 23:18 UTC
Message-ID: <4E6FE4DF.7060200@ramsay1.demon.co.uk>
In-Reply-To: <CAGdFq_ifU2WWbCpRY_EFY=_hwwtFs0eqMhJ7sRoUhrivABFoFw@mail.gmail.com>

Sverre Rabbelier wrote:

Show 18 quoted lines

> Heya,
> 
> On Sun, Sep 11, 2011 at 21:26, Ramsay Jones <ramsay@ramsay1.demon.co.uk> wrote:
>>    compat/obstack.c:399:1: error: symbol 'print_and_abort' redeclared with \
>>        different type (originally declared at compat/obstack.c:95) \
>>        - different modifiers
> 
>> @@ -395,7 +395,6 @@ _obstack_memory_used (struct obstack *h)
>>  # endif
>>
>>  static void
>> -__attribute__ ((noreturn))
>>  print_and_abort (void)
>>  {
>>   /* Don't change any of these strings.  Yes, it would be possible to add
> 
> Wouldn't the better solution be to add noreturn to the declaration at
> compat/obstack.c:95?

Hmm, well ... maybe; it is at least debatable. But I decided no! :-D

First, although I would not dismiss the possibility of some optimization of the code of print_and_abort() (the *callee*), the main benefit of the noreturn attribute should in fact be at the call sites (ie the *caller*). So, yes, in general, the declaration of the function should have the noreturn attribute applied, in addition to the definition, in order to allow the compiler to apply some optimizations to the call sites.

[Note, also, that we should use the NORETURN and NORETURN_PTR macros.]

In this case, however, there are no (direct) call sites. This function would only be called indirectly via the 'obstack_alloc_failed_handler' function pointer. So, this would require the use of NORETURN_PTR on that function pointer. In order to keep both the compiler(s) and sparse happy, the required change would look like the diff given at the end of this mail.

This would work fine, and I would happily change the patch to include this if it is deemed the better approach. However, I looked at the call sites in _obstack_begin[_1](), and _obstack_newchunck() and could not see any great opportunity for optimizing the code ... so I decided to go for the simpler patch ...

ATB, Ramsay Jones

-- >8 --

diff --git a/compat/obstack.c b/compat/obstack.c
index a89ab5b..2029b8f 100644
--- a/compat/obstack.c
+++ b/compat/obstack.c
@@ -92,8 +92,8 @@ enum
    abort gracefully or use longjump - but shouldn't return.  This
    variable by default points to the internal function
    `print_and_abort'.  */
-static void print_and_abort (void);
-void (*obstack_alloc_failed_handler) (void) = print_and_abort;
+static void NORETURN print_and_abort (void);
+NORETURN_PTR void (*obstack_alloc_failed_handler) (void) = print_and_abort;
 
 # ifdef _LIBC
 #  if SHLIB_COMPAT (libc, GLIBC_2_0, GLIBC_2_3_4)
@@ -395,7 +395,7 @@ _obstack_memory_used (struct obstack *h)
 # endif
 
 static void
-__attribute__ ((noreturn))
+NORETURN
 print_and_abort (void)
 {
   /* Don't change any of these strings.  Yes, it would be possible to add
diff --git a/compat/obstack.h b/compat/obstack.h
index d178bd6..122f93f 100644
--- a/compat/obstack.h
+++ b/compat/obstack.h
@@ -194,7 +194,7 @@ void obstack_free (struct obstack *, void *);
    more memory.  This can be set to a user defined function which
    should either abort gracefully or use longjump - but shouldn't
    return.  The default action is to print a message and abort.  */
-extern void (*obstack_alloc_failed_handler) (void);
+extern NORETURN_PTR void (*obstack_alloc_failed_handler) (void);
 
 /* Pointer to beginning of object being allocated or to be allocated next.
    Note that this might not be the final address of the object